[FFmpeg-devel] [PATCH] lavc/aarch64: add sbrdsp neon implementation

Matthieu Bouron matthieu.bouron at gmail.com
Tue Jun 13 16:54:06 EEST 2017


On Fri, Jun 02, 2017 at 01:59:00PM +0200, Matthieu Bouron wrote:
> ---
> 
> Hello,
> 
> The following patch adds an aarch64 neon implementation of the sbrdsp (tested
> on an Odroid-C2). It hasn't been benchmarked yet and it lacks the
> hf_apply_noise{0,1,2,3} functions (which will be added later).

New patch attached adding the missing hf_apply_noise{0,1,2,3} functions.

Here are the performance results on an Odroid-C2:

sbr_apply_noise_0_c: 1690.6
sbr_apply_noise_0_neon: 1498.4
sbr_apply_noise_1_c: 1689.4
sbr_apply_noise_1_neon: 1504.2
sbr_apply_noise_2_c: 1690.2
sbr_apply_noise_2_neon: 1498.4
sbr_apply_noise_3_c: 1694.8
sbr_apply_noise_3_neon: 1504.0
sbr_autocorrelate_c: 644.0
sbr_autocorrelate_neon: 420.0
sbr_deint_bfly_c: 1107.6
sbr_deint_bfly_neon: 291.6
sbr_deint_neg_c: 210.4
sbr_deint_neg_neon: 107.4
sbr_filt_c: 2118.2
sbr_filt_neon: 1219.5
sbr_gen_c: 6071.2
sbr_gen_neon: 3251.2
sbr_neg_odd_64_c: 70.0
sbr_neg_odd_64_neon: 64.7
sbr_post_shuffle_c: 163.0
sbr_post_shuffle_neon: 107.7
sbr_pre_shuffle_c: 120.5
sbr_pre_shuffle_neon: 110.7
sbr_sum64x5_c: 1361.6
sbr_sum64x5_neon: 435.4

-- 
Matthieu B.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-lavc-aarch64-add-sbrdsp-neon-implementation.patch
Type: text/x-diff
Size: 18055 bytes
Desc: not available
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20170613/048c4167/attachment.patch>


More information about the ffmpeg-devel mailing list